Ok,
Clients would write to logs .. servers would replay logs .. as a server controls it's own unique data-store, there is no conflict of interest / sync problem.
>Imagine a case where there is 2 clients changing the same AFR file.
Both would write to the same shared log file which would be stored ON th AFR.
When a server comes back up, it would replay the logs BEFORE becoming available to clients and would only become available once in-sync ...
Gareth.
----- Original Message -----
step 3.: "Thierry EXCOFFIER" <exco@xxxxxxxxxxxxxxxxxxxx>
To: gluster-devel@xxxxxxxxxx
Sent: 17 January 2008 12:37:43 o'clock (GMT) Europe/London
Subject: Re: Helping AFR ( Checksumming or Logging )
Angel wrote:
> AFR (LOGTRANS(A),LOGTRANS(B))
>
> when A FAILS LOGTRANS(A) get signaled when child dies and start logging.
>
> on child recovery LOGTRANS(A) get signaled againg and replays TRANS..
Imagine a case where there is 2 clients changing the same AFR file.
If each client replay its log without synchronizinq with the
other client the final file may be not in the same state on A and B
after the replay.
My 2 cents.
PS : To speed up AFS synchronisation, I prefer the idea of
a 'split' translator that take apart big files in chunks :
bigfile/
chunk1
chunk2
chunk3
...
So AFR synchronize only the modified chunk.
By the way, an 'unsplit' translator may be needed at the top
(The 'unsplit' translator may be useful to store files bigger
than the maximum size allowed by the file system)
--
Thierry.EXCOFFIER@@liris.univ-lyon1.fr// GO: Oriental Strategy Game
LIRIS/UFR Informatique, bat710 //DYLAN: Programming Language
Universite Claude Bernard LYON 1 __// UNIX: Standard Operating System
43 Bd du 11 novembre 1918 //
69622 VILLEURBANNE, FRANCE //WWW: http://www710.univ-lyon1.fr/~exco
_______________________________________________
Gluster-devel mailing list
Gluster-devel@xxxxxxxxxx
http://lists.nongnu.org/mailman/listinfo/gluster-devel
